In "lpm-profile" mode,
- It uses a combination of TCAMs and SRAMs in PFE to implement longest prefix match
- Its algorithm groups entries with common prefixes together
- TCAM stores the root prefix and SRAM stores the list of prefixes associated with the root prefix
- The number of entries in this secondary list determine the total number of prefixes that can be supported.
- Depending on the pattern of incoming prefixes, the scale number could vary
=> There is no fixed max scale number for prefixes
You can see max numbers per prefix types in the output of "show pfe route summary hw" command
and case by case, you may see the number of prefixes in "Used" column exceeds the number of prefixes in "Max" column.
- Numbers in MAX column are provided by PFE chip vendor and those are minimum guaranteed numbers
- Because the reason explained above, more than the MAX number of prefixes can be programmed like example below.
root@QFX5K> show pfe route summary hw
Profile active: lpm-profile
Type Max Used Free % free
----------------------------------------------------
IPv4 Host 8192 822 5558 67.85
IPv4 LPM 131072 136855 0 0.00
IPv4 Mcast 4096 0 2779 67.85
IPv6 Host 4096 906 2779 67.85
IPv6 LPM(< 64) 16384 17781 0 0.00
IPv6 LPM(> 64) 16384 11 0 0.00
IPv6 Mcast 2048 0 1390 67.87
There is no other way to monitor the utilization of TCAM and SRAM. Only way to know how much prefixes can be programmed is trying to programm
the exact same pattern of the prefixes you want to use until you see table full error.